Decision Cards

How I work with AI: it asks instead of guessing. Each call that needs me comes as one short question on a card, and I answer Yes, No, or in my own words.

Try it

These cards are made up, about the same made-up family as my family history demo. Answer them the way I answer mine.

5 waiting · 0 answered · 0 in your own words

  1. Family history Waiting

    Clara Halvorsen’s baptism says she was born in 1884; a census taken years later says 1885. Use 1884?

    Leans Yes The baptism was written the week she was born. The census came later, from memory.

  2. Family history Can’t be undone Waiting

    Delete the 12 duplicate scans of Ole Halvorsen’s letters? Each one is an exact copy, and the originals stay.

    Leans Yes They are exact copies, and they make the letters look like 24.

  3. Family history Goes out in your name Waiting

    Send this thank-you to the archivist at the county archive?

    The message, word for word

    Hello, and thank you for finding the 1849 marriage record of Anders Halvorsen. It answered a question the family has asked for years. With thanks.

    Leans Yes She went out of her way, and a short thank-you keeps the door open.

  4. Family history Waiting

    Henry Halvorsen signed his name Halvorson after he moved to Minnesota. Keep Halvorsen on his card?

    No lean Both are right, for different years. This one is your call.

  5. Family history Waiting

    Someone else’s online tree gives Lars Halvorsen a third son, Nils. Add him?

    Leans No No record shows a Nils yet. A search for his baptism should come first.

AI is quick to act, and just as quick to guess. Across my projects, getting AI to do the work was the easy part. The hard part was making sure the calls that are mine came to me, and stayed made.

At first that meant long chats with Claude. At this scale they were hard to follow, the answers were too long to act on, and a later pass could undo something I’d already decided. So I turned it around: when the AI reaches a call it shouldn’t make alone, it stops and asks me on a card.

One card, one question

Each card is one short question I can answer in seconds. It says which way the AI leans and why, so I’m checking a judgment instead of starting from scratch.

  • Yes or No

    Most cards take one tap. The work picks up from my answer, and the result comes back on the same card.

  • Or my own words

    When neither fits, I write what I want instead, and the AI follows it. Anything that needs me again comes back as a new card.

  • Calls that stay made

    Once I’ve answered, the call is settled. A later pass can’t quietly reopen it.

The rules it runs on

Three rules, in plain words. Every card follows them, and so does all the work that comes out of one.

  • Nothing goes out in my name without my yes

    Every message waits, word for word, until I say yes. The AI can write the draft, but it can’t send it alone.

  • A human approves anything that can’t be undone

    If a step can’t be taken back, a person says yes first. The AI can get it ready, but it can’t finish it alone.

  • I can see and undo everything

    Every answer, and what came of it, stays on the page. Anything that can be undone, I can undo; the rest needed my yes first.

So far

I’ve answered 122 cards, 39 of them in my own words. Yes and No cover most calls, but not all of them, which is why every card also takes my own words.

Where I use it

I use the same cards across my projects: wherever AI does the work and a call is mine to make.

What changed

  • Long chats became one-line questions I can answer in seconds
  • Calls I’ve made stay made: a later pass can’t undo them
  • The AI does the work between cards, and I spend my time on the calls
  • Nothing goes out in my name, and nothing permanent happens, without a yes from me

How can I help you?

Contact me